Hyperscalers Hedge: Build Models In‑House

Despite deep partnerships with labs like OpenAI, Microsoft says it will train its own frontier models on much larger GPU clusters. This dual track—partner and vertically integrate—lets hyperscalers control capability roadmaps and bargaining power while still using third‑party models when convenient. — It signals consolidation and a harder competitive race in AI that will shape model access, antitrust debates, and energy demand.